Multicomponent biomarker approach improves the accuracy of diagnostic biomarkers for psoriasis vulgaris

Abstract

Accurate biomarker-based diagnosis of psoriasis vulgaris has remained a challenge; no reliable disease-specific biomarkers have yet been identified. There are several different chronic inflammatory skin diseases that can present similar clinical and dermoscopy features to psoriasis vulgaris, making accurate diagnosis more difficult. Both literature-based and data-driven selection of biomarker was conducted to select candidates for a multicomponent biomarker for psoriasis vulgaris. Support vector machine-based classification models were trained using gene expression data from locally recruited patients and validated on 7 public datasets, which included gene expression data of other inflammatory skin diseases in addition to psoriasis vulgaris. The resulting accuracy of the best classification model based on the expression levels of 4 genes (IL36G, CCL27, NOS2 and C10orf99) was 96.4%, outperforming classification based on other marker gene combinations, which were more affected by variability in gene expression profiles between different datasets and patient groups. This approach has the potential to fill the void of clinically applicable diagnostic biomarkers for psoriasis vulgaris and other inflammatory skin diseases.